Not all Sparta histograms stats are reset after warmup

[benoy-sifive]

I believe there is a stat bug in Sparta for histograms. What I believe is happening is that the utilization totals are not being reset once we hit the end of the warmup window. You can see that here:

            entries_utilization_UF = 0
            entries_utilization_UF_probability = 0
            entries_utilization_OF = 0
            entries_utilization_OF_probability = 0
            entries_utilization_total = 10263 <== Should have been reset
            entries_utilization_max = 11 <== Should have been reset
            entries_utilization_weighted_avg = 0
            entries_utilization_weighted_nonzero_avg = nan

[benoy-sifive]

Note:

*_total stat uses cycle counters. Are all cycle counters automatically reset at warmup? e.g. if counting then stopCounting()...startCount()
*_max stat uses COUNT_LATEST. Not sure if these need to have a special check for reseting after warmup

[ghost]

Yeah, that's weird. Will take a look

[ghost]

Can you tell me which resource you're using that has this histogram? Array, Buffer, Queue, CircularBuffer, etc?

[benoy-sifive]

Buffer

[ghost]

Well, one "mystery" somewhat solved -- entries_utilization_max is a COUNT_LATEST and it updated independent of a report. The counter is updated when the the previously recorded value is less than the new value, but the report will never know that. Not sure how to make that one work...

[ghost]

Hmmm... I cannot reproduce the issue with entries_utilization_total being incorrect. I added a test to Buffer_test generating 2 reports, one without warmup and the other with warmup. The test ran for 20 cycles total.

Without warmup (report 1):

    buf_const_test_utilization_total = 20

With warmup (report 2):

    buf_const_test_utilization_total = 10

buf_const_test_utilization_max is of course still wrong...

[ghost]

I can't reproduce utilization total issue, but I'll keep this open in case inspiration hits me to fix the max problem.